IBM(NYSE: IBM) will start offering Nvidia's (NVDA 2.03%) Tesla M60 GPU accelerators to its cloud-based enterprise clients. IBM states that installing the high-end GPUs will enable companies to "deploy fewer, more powerful cloud servers while churning through complex jobs faster." These jobs include compute-intensive workloads from graphics, analytics, energy exploration, and deep learning virtual applications.
This isn't the first time IBM teamed up with Nvidia. Big Blue already offers Nvidia's Tesla K80 and K10 GPUs to its cloud-based clients. Last year, IBM claimed that combining the K80 with its Power-based chips enabled its Watson AI platform to answer queries 1.7 times faster. That same year, IBM partnered with Nvidia and the Department of Energy to install its Power8 CPUs alongside Tesla P100 GPUs in two new supercomputers by 2018.
Let's examine how these collaborations are helping both companies widen their defensive moats.
Nvidia's top-tier Tesla P100. Image source: Nvidia.
Why IBM needs Nvidia
Partnering with Nvidia helps IBM in two ways. First, it strengthens its Open Power initiative, which "open sources" its Power-based server designs to third-party vendors. IBM controls a tiny sliver of the data center market compared to Intel (INTC 4.02%), but pairing Nvidia GPUs with its Power CPUs could prove to be a potent combination for next-gen applications like cloud-based apps, analytics, and AI.
That's because machine learning applications generally run faster on GPUs than CPUs, and Nvidia is the market leader in high-end GPUs. Nvidia claims that the Tesla K80 GPU is "two to five times" faster than Intel's Xeon Phi 7120 CPU at running "key science applications." Higher sales of Power CPUs into its first and third-party servers could boost IBM's system (systems hardware and operating systems software) sales, which fell 22% annually last quarter and accounted for 9% of its total revenue.
Second, customers who buy Power-based hardware could be more likely to buy additional services from IBM's higher growth "strategic imperatives" of cloud, analytics, and engagement businesses. Revenue from those units rose 14% annually last quarter and accounted for 37% of the company's top line. IBM hopes that growth in hardware and strategic imperatives could help it finally break its 16 straight quarters of annual revenue declines.
Why Nvidia needs IBM
For Nvidia, Tesla/GRID GPUs for cloud computing can help it grow its business beyond high-end GPUs for PCs. Last quarter, Nvidia's data center revenue (which includes Tesla/GRID), rose 63% annually and accounted for 11% of the company's top line. Nvidia attributed that growth -- which outpaced 17% growth in its core gaming GPU business and 4% growth in professional visualization -- to "strong demand for GPU acceleration related to deep learning." Tesla/GRID GPUs also connect to Nvidia's Tegra CPUs, which are being installed in infotainment systems and its end-to-end solutions for autonomous cars.
Nvidia's alliance with IBM also complements its partnership with Facebook (FB 2.04%), which installed Tesla M40 GPUs in its Big Sur computing platform for AI tasks last year. Like IBM, Facebook open sourced Big Sur's design to other companies, which could boost marketwide demand for Tesla GPUs.
Facebook's Nvidia-powered Big Sur platform. Image source: Facebook.
But looking ahead, Nvidia's Tesla faces big challengers in the data center business. Alphabet's Google recently announced the development of its own machine learning chip, the TPU (Tensor Processing Unit), which is already powering some of its data centers. Google has claimed that TPUs could be more efficient than GPUs in AI applications. That's why Nvidia needs major allies like IBM and Facebook in its corner.
But it's all bad news for Intel
IBM and Facebook's partnership with Nvidia and Google's development of first-party chips indicate that the data center market is rapidly evolving. That's good news for Nvidia, which has established itself as the "best in breed" player in graphics processing.
But it's bad news for Intel, which lacks Nvidia's high-end GPU muscle and will likely need to rely more on the FPGA (field programmable gate array) tech it gained through its acquisition of Altera. Intel still controls around 99% of the data center market, but the tables could quickly turn if Internet heavyweights like Facebook and Google start ordering fewer Intel chips.
The key takeaway
IBM and Nvidia's new partnership is a win-win deal for both companies, but it arguably helps Nvidia more than IBM. That's because Tesla sales account for a higher percentage of Nvidia's overall revenue, and are rising at a much faster rate than IBM's systems or strategic imperatives revenue. For IBM, partnering with Nvidia is a defensive move which shrewdly widens its moat against Intel.
